Real-time Track and Anomaly Detection in Complex Railway Environment

Abstract

Autonomous vehicles, these days have been gaining a lot of interest and it seems to promise a safer, more reliable world. Autonomous cars on one hand have bloomed into commercial production, but on the other hand, autonomous trains in particular have not yet been in the limelight. The paper builds around the premise that in unmanned railway technology, perceiving the driving environment in front of the train and identifying the potential safety threats are critical issues. In response, we propose a way based on computer vision to detect the railway tracks in real-time which can be used for safety and automation purposes. Followed by anomaly detection using deep learning based object detection algorithm. We experimentally show that track extracted has good continuity and low noise, and the probable obstacles also get appropriately detected.